Amazon Infographics in Minutes

Minutes, not seconds, and the honest unit is "two background jobs" rather than "one click". Pasting an ASIN is instant. The research pass makes fourteen sequenced network calls before it writes a word. The image set then renders every script at once, so a six-image set takes roughly as long as one image. Our own API waits ten minutes by default for a run to finish and gives up at fifteen. That is the real envelope.

What the clock is actually measuring

Two jobs, not one spinner. Research scrapes and thinks; image generation renders. They are queued separately, run in the background, and you can close the tab — the dashboard polls and picks up where things stand. Nothing here depends on you sitting still, which is the part most "in minutes" claims quietly rely on.

Concurrency is capped per job type. Research runs four jobs at once behind a rotating residential proxy and exactly one without it, because parallel scrapes from a single IP walk into Amazon's captcha wall. Image generation runs two at once. The wall-clock answer for one product is therefore not the answer for five, and any page quoting a single number is not describing a queue.

Step one: research, fourteen steps

The research job fetches your listing, resolves its deepest category, pulls that category's best-seller list, then fetches up to ten competitor listings one at a time, with a deliberate 900-millisecond pause before each. That alone is about nine seconds of built-in politeness, plus twelve page loads through the proxy. The last step is a single model call that reads all of those bullets, specs, prices, ratings and reviews and returns the brief: pain points tagged by whether they came from your reviews or a competitor's, USPs grounded in that evidence, a positioning angle, a colour and type style, and six or seven image scripts.

That last call is the slowest single thing in the product. The model reasons before answering and the JSON alone runs several thousand tokens, so the output cap sits at 64,000 tokens. If the brief still runs out of room, the job retries once on a tighter payload — five competitors instead of ten — which is cheaper than failing and costs a second model call in wall time. Research costs 0.1 credit either way, and a blocked scrape fails and refunds itself.

Step two: the set renders in parallel

Once the brief exists, generating the set issues every script to the image model concurrently rather than in sequence. A seven-image set is therefore about as fast as the slowest single image, not seven times one image. Images that fail inside the run are refunded individually; the rest of the set still lands.

One caveat with a number attached: the Minimal style mode uses a flat style tile as its visual anchor, and that tile is generated before the set fans out, adding one call to the front of the job. When you have a brand kit, the browser renders that tile locally from your exact hexes, fonts and logo — free, and no extra wait. Vivid, Studio and Technical paint their own backgrounds and skip the tile entirely.

The number we are willing to put in writing

Our public API is the most honest source on duration, because it had to pick a timeout. POST /v1/infographics chains research and images into one run, and wait: true holds the connection open until the images land: ten minutes by default, never more than fifteen. Those figures were chosen to cover a cold ASIN through to six finished images. Treat ten minutes as the design envelope for a run from nothing, and less when research already exists — an ASIN researched earlier skips straight to the images.

Editing is where "minutes" is actually true

The generation chain is minutes; the edit loop is seconds, and that is the difference that matters after the first version. Layerize runs one model call to lift baked-in text into editable objects and repaint the plate beneath, at 0.5 credit. After that, changing a headline is typing — no queue, no credit, no regeneration. A spot edit is 0.5 credit, translating a finished image is 1.

Export costs nothing and waits on nothing: the canvas renders to PNG at 2160×2160 in your browser, or to SVG with the text still live. Worth knowing for Amazon: generated gallery images come back at 1024 px, which clears Amazon's 1,000 px floor for zoom, while the 1,600 px or more that Amazon recommends comes from the Studio export.

How to make it fast on purpose

Research everything you might build before deciding what to build. At 0.1 credit per ASIN, running four at a time in the background, it removes the slowest step from the moment you are actually in a hurry: when a set is needed the brief already exists and you are only waiting on pixels. Choose Nano Banana 2 or GPT-5 Image Mini at 1 credit when speed matters more than the last increment of quality. And do not start five sets to save time — two render and three queue.

Who this speed suits, and who it does not

If you need one image in thirty seconds, use a prompt-and-render tool; nothing about a fourteen-step research pass will beat that, and we would rather say so now than lose the argument later. If you need six images that argue a coherent case before a deadline this afternoon, this is a good trade: the thinking takes a few minutes instead of an afternoon, and each revision takes a minute instead of a designer's day. If your deadline is next month, speed is not why you would pick any tool here.

Common questions

How long does it take to make an Amazon infographic in Graflio?

Minutes, in two background stages. Research makes twelve page fetches plus one long model call across fourteen steps; the image set then issues all six or seven scripts at once, so it takes about as long as a single image. Our API holds a waiting connection for ten minutes by default and fifteen at most, which is the intended envelope for a full run from a cold ASIN.

Can I generate several products at the same time?

Yes, but they queue. Research runs four jobs concurrently behind a rotating proxy and one without it; image generation runs two jobs at a time. Starting five image sets does not make five sets appear five times faster — two render and three wait.

What is instant and what is not?

Instant and free: opening an image in the Studio, retyping or restyling text after layerize, and exporting to PNG, SVG or PSD, all rendered locally. Not instant: the first research pass, any image generation, and the AI edit actions — layerize 0.5 credit, spot edit 0.5, translating an image 1.

What happens if it takes too long or fails?

Jobs refund themselves. A failed research job returns its 0.1 credit; individual images that fail inside a set are refunded while the rest of the set completes; a job lost to a server restart is failed and refunded on boot. If the brief runs out of output room, the job retries once automatically on a smaller set of competitors rather than making you re-scrape.

Is it faster the second time?

Considerably. A listing that has already been researched keeps its brief on the product, so regenerating a set — in another style mode, or with a different model — skips straight to the images. Through the API, a run on an already-researched ASIN reuses that brief unless you explicitly ask for a fresh one.